Renting AI vs Owning It: The Build vs Buy Arithmetic

MyCase Advanced is published at $130 per user per month annually: twenty seats run $31,200 a year. Smith.ai's Pro receptionist plan is published at $2,100 a month for 300 calls, or $25,200. Two line items: $56,400 a year, $4,700 a month. Month 10 against a $45,000 build, month 26 against $120,000, month 39 against $180,000. Add MyCase Accounting at $39 a seat and it is $65,760: month 9, month 22, month 33.

Smith.ai's Pro plan is $2,100 a month and includes 300 calls. At 500 calls a month, 200 overage calls at the published $8.50 add $1,700. The month is $3,800; the year is $45,600. Answering the phone, one vendor, every figure on smith.ai/pricing, crosses a $45,000 build in month 12, $120,000 in month 32, $180,000 in month 48. Five years of it runs $228,000.

A 20-attorney firm on Harvey cannot be modeled. harvey.ai/pricing returns 404, and the estimates that exist are scoped to firms of 50 to 200 attorneys with a reported 20-seat minimum. The firm cannot learn the price without a sales call, and the 404 is the receipt.

CoCounsel Essentials fails differently. Thomson Reuters publishes nothing, and the only figure in circulation is a third-party estimate from the aggregator costbench.com, roughly $293 per user. That estimate does not say whether it is monthly or annual, and the readings differ twelvefold: $17,580 a year against $1,465 for five seats. CoCounsel also cannot be bought without a Westlaw base, whose price is likewise unpublished.

A 25-person firm on Canopy Premium at the published $149 a seat pays $44,700 a year, the published Tax Resolution add-on at $50 adds $15,000, and Jetpack Workflow Premium at $50 adds another $15,000. Three published line items, no estimates: $74,700 a year, $6,225 a month. Month 8 against $45,000, month 20 against $120,000, month 29 against $180,000. A firm that size is typically steered to quote-only Enterprise pricing, so the published rate is likely a floor.

Tax prep is a separate line. Lacerte Pay-Per-Return publishes a $599 Fast Path license, $105 per individual return with one state and $148 per business return. At 500 individual and 100 business returns that is $67,899 a year: month 8, month 22, month 32. The return mix is our labeled assumption; the rates are Intuit's. Lacerte Unlimited, the edition that firm would buy, is quote-gated.

CCH Axcess returns 403 on all three product pages and on the Wolters Kluwer terms of use. UltraTax CS redirects its pricing portal to an account sign-in, so price and license are visible only to existing customers.

A 15-user shop on Jobber Grow prepaid annual is published at $320 a month, $3,840 a year, and the three published add-ons stack on top: Marketing Suite $948, Receptionist $348, Pipeline $588. Fully loaded, $5,724 a year, $477 a month: month 95 against a $45,000 build, month 252 against $120,000.

Housecall Pro is computable only in part. MAX is published at $299 a month annual with extra users at $35, so the base is $3,588 a year and each added seat $420, but the included-seat count was not captured, so a 25-technician total cannot be computed honestly. Smith.ai and Dialzara cannot be compared to each other either, one billing per call and one per minute with no published average call length. Both publish full rates, more transparent than every legal-AI and insurance vendor here combined.

Read the Smith.ai row twice: a vendor holding law-firm call recordings with no public statement about what it does with them. Legora is the other tell, promising "Zero AI training on your data" on a homepage while its privacy policy excludes Subscriber Content.

Botkeeper belongs beside them: Derived Data is kept for "any business purpose" in perpetuity. Vic.ai and Blue J are the cleanest receipts for the sentence operators rarely hear: you can own your data and not own what it built.

Three caveats, in the open rather than a footnote. The EZLynx clause says "to the Sites" and sits in the website Terms of Use, so whether it reaches data entered inside the product is a real ambiguity counsel would raise; it does not support the paraphrase that EZLynx owns your book of business, and must not be attributed to Applied Epic, a different product under the same parent. The Vertafore language is from search-index excerpts of a non-extractable PDF, not a verbatim read.

Workiz belongs here too: it "may permanently delete Pro's Pro Content." Jobber is the cheapest published stack here and carries its broadest license grant; price alone shows only the first.

The claim that vendors never give your data back is false, and this page's own dataset would refute it. The defensible claim is narrower and worse: the exit exists only if you know the window, and the window is short, undisclosed or absent more often than not.

Dext allows ten days from termination to make a written request. Harvey deletes within thirty. MyCase gives thirty days read-only. LexisNexis retains not less than sixty. ServiceTitan is not obligated beyond sixty, and its export is "commercially reasonable efforts" rather than a guarantee. Clio reportedly allows ninety and Jetpack Workflow may erase everything within ninety. Twenty-four of the thirty-two have no export mechanism at all, and a firm that cancels Dext and goes quiet loses everything on day eleven.

The strongest published data terms here belong to competitors: LexisNexis a no-training warranty and a named sixty-day export window, ServiceTitan the clearest AI-training clause surveyed, Dext the most concrete export mechanism in the file. Naming them is what makes the rest of this page worth reading.

The counterweight is not hypothetical. In February 2026 Botkeeper announced an abrupt shutdown after eleven years and roughly $90 million raised, per Accounting Today, Inside Public Accounting and CPA Practice Advisor. Xendoo acquired the platform within two weeks and relaunched it as Botkeeper Infinite: those firms kept their workflow but inherited a counterparty they never chose.

It depends on what you already spend. A 25-person CPA firm on three published subscriptions pays $74,700 a year and passes a $45,000 owned build in month 8. A 15-user home-services shop on a loaded Jobber stack pays $5,724 and does not pass one until month 95.

On published rates, a 20-attorney firm on MyCase Advanced and Smith.ai's Pro plan pays $56,400 a year, or $65,760 with MyCase Accounting. A 25-person CPA firm on Canopy Premium plus Tax Resolution plus Jetpack Premium pays $74,700. Lacerte Pay-Per-Return at 600 returns is $67,899. For most legal-AI tools and all five insurance vendors, nothing is published.

When your spend is small enough that the crossover sits years out. A 25-person CPA firm on the leanest published stack pays $148,500 over five years, so a $180,000 build loses outright. A 15-user home-services shop pays $28,620, $16,380 less than our smallest commission. Blue J's Individual plan at $1,498 a year is right for a sole practitioner.

Usually you own the data and not what the data built. Vic.ai's agreement says it "exclusively owns Derived Data, including all intellectual property rights therein." Blue J's grants permission "to review all inputs made by Users into the Product and corresponding outputs," no opt-out. Harvey, LexisNexis, CoCounsel and Spellbook do commit to no training, explicitly.

It depends on a window you have probably not read. Dext requires a written request within ten days. Harvey deletes within thirty. MyCase gives thirty days read-only. LexisNexis retains not less than sixty, ServiceTitan is not obligated past sixty, Jetpack Workflow may erase everything within ninety. Twenty-four of thirty-two have no export mechanism at all.

We will not guess at motive, but the pattern is reportable. Nine of thirty-two publish a usable rate card. Zero of the six legal-AI tools do. Zero of the five insurance vendors do: Quandri 404, Applied Epic 404, Vertafore contact-us, Ushur nothing, EZLynx "no single flat rate." CCH Axcess returns 403 and UltraTax CS redirects to a login wall. Stacking is easiest to miss when each line arrives through a sales call.